Multiobjective Eco-Driving and Energy Management Coordination Optimization Strategy for FCVs Based on Deep Reinforcement Learning in Connected Scenarios

With the support of vehicle-to-everything technology, the eco-driving and energy management coordination optimization can further enhance the performance of fuel cell vehicles. However, the complexity of multi-dimensional spatiotemporal information and the strong coupling of the vehicle-road system pose significant challenges to the real-time multi-objective energy management of fuel cell vehicles (FCVs) in intersection scenarios. This article proposes a dual-agent intelligent energy management strategy for FCVs based on the twin delayed deep deterministic policy gradient algorithm. Specifically, a vehicle-road coupled eco-driving agent is designed to achieve four-dimensional balance of energy efficiency, comfort, safety, and traffic efficiency. Look-ahead traffic light reward is designed to improve the smoothness of vehicle passing through traffic lights. The task termination conditions are introduced to improve training efficiency and driving safety. Additionally, the spatiotemporal information-integrated energy management agent is designed, where the multi-objective reward function helps improve energy efficiency and enhance fuel cell durability. The simulation results demonstrate that the proposed strategy can decrease fuel consumption, ensure safe driving, and have the potential to improve fuel cell durability. The proposed strategy achieves smoother acceleration profiles. Additionally, the hardware-in-the-loop experiment within the cloud-vehicle-road framework is conducted.